fix(planning): hand-written plan-approval reconfirm row does not round-trip through ledger import #5534

Follow-up to #4611 (PR #5470 leaves it out of scope).

Problem

On the terminal path, /planning:plan tells the session to hand-write the reconfirm resolution of a superseded-by-plan row as reconfirmed at plan approval: <new>; was: <old>:

• plugins/planning/skills/plan/SKILL.md:275
• plugins/planning/skills/interview/context/loop.md:238

Before PR #5470, export-ledger wrote that text for a page accept on a superseded row; the named-field grammar replaces it, so only the hand-written path still produces it. import-ledger has no reader for it and treats the whole string as the row's own text, so the new and old answers do not come back as a structured pair. Pinned by the superseded-accept fixture in plugins/planning/surface/tests/fixtures/ledger-legacy/: the imported terminal decision is own with text equal to the full string, commits empty, and no was.

Scenario (c) of #4611 (a superseded row reconfirmed by an accept keeps its was answer as structure) therefore round-trips on the page path once #5470 merges and is still lost on the hand-written terminal path.

Options for triage

1. Change both skill texts to hand-write the named-field form the exporter writes for the same state, so import restores it.
2. Teach the importer to read the hand-written reconfirmed at plan approval: <new>; was: <old> form into the same structured state as option 1.
3. Both, with the importer reader kept as a legacy grammar.

The importer must keep accepting the current text either way, because ledgers with it already exist.

Acceptance criteria

• A ledger row a session hand-writes for a reconfirmed superseded row imports to the same question state as the page path's accept on a superseded row (decision, was, proposal pair).
• The superseded-accept legacy fixture still imports unchanged, or its expected state is updated with the reason stated in the PR.
• plan/SKILL.md and interview/context/loop.md describe one form; interview-defenses.test.sh digest re-pinned if loop.md changes.
• A test round-trips the hand-written form.

Out of scope

The page export grammar itself (#4611) and the two held-row losses named in PR #5470.